Features
Every named feature of the platform — the hardware, engines, app modules, and record layer a customer gets. Each card is one feature; follow a section’s deep-dive link for the full story.
Capture hardware
Section titled “Capture hardware”The input devices that record work as it happens. Deep dive: Iris · Oracle platform
Iris Clip-On
A clip-on camera for the technician's own eyewear — first-person video and audio from their exact point of view, with on-board edge AI.
Oracle Universal Core
One modular camera unit — Ultra-HD low-light sensor, studio-grade mic, wireless — for every scenario.
Magnetic SNAP! bases
The Core snaps onto chest pendant, stand, ceiling, and arm mounts — continuous capture, no charging gaps.
Audio Pendant
All-day wearable audio with noise cancellation, a hardware privacy toggle, and hot-swap battery.
Voice & tap controls
"Iris, start / stop / mark / private" plus tap-to-bookmark — hands never leave the work.
Robot-grade time sync
±1 ms PTP timestamps, ROS-friendly; first- and third-person views fuse into one session.
Privacy & security
Section titled “Privacy & security”The controls that get the system past EHS, IRB, and IP review. Deep dive: Enterprise Security & Privacy
Edge redaction
Faces and badges blurred on-device before anything transmits; raw feeds never auto-save to cloud.
Hardware privacy controls
Physical toggles, recording-state LEDs, and a remote Force Privacy Mode kill switch.
Isolated data silos
Per-customer graph and vector stores, AES-256 at rest, single-tenant and on-prem options.
Intelligence engines
Section titled “Intelligence engines”How raw capture becomes verified truth. Deep dive: System Architecture
Asset Fingerprinting
Parts identified on sight — an image is as good as a barcode. No scanning ritual, no manual entry.
The Session engine
Video, audio, Slack, email, and CAD saves normalize into one AI-drafted event format.
Human-in-the-loop gate
AI drafts, an engineer approves — nothing mutates the permanent record without sign-off.
Unknown-token handling
Unrecognized parts are flagged on the timeline with everything the system saw — never guessed.
SKU taxonomy + CAD mapping
One strict part vocabulary linking vision, CAD, SOPs, and inventory to the same physical part.
Three-phase BOM-drift detection
Drift caught at design time, in the act on the floor, and in the post-assembly audit.
Anomaly webhooks
Every alert carries severity, expected vs. detected part, and a link to the exact video clip.
Agentic RAG queries
Ask in plain language across every store; get the time-stamped clip that answers it.
The apps — Outpost & Hub
Section titled “The apps — Outpost & Hub”The technician's command center on iOS and in the browser. Deep dive: Outpost · The Hub
Conversational Dashboard
Pull up builds, SOPs, or inventory by asking — on a drag-and-drop grid of live widgets.
Quick Actions palette
One gesture — logo tap or ⌘K — to start SOPs, recordings, or inventory checks mid-task.
Interactive Timeline
Every event deep-links to its source video or document; verify, edit, or ignore inline.
Living, multimodal SOPs
Auto-generated guides pairing each step with video, live part status, and an AI copilot.
Autonomous Inventory
Counts deduct from what cameras observe during a build — zero manual entry.
Sessions Management
One inbox for every input stream — recordings, Slack, email, field notes — awaiting review.
Live Tech Support
A remote expert sees the technician's live stream while the AI checks each step in real time.
Component Registry
A browsable catalog of the full SKU taxonomy with a live validator.
Audit Log
BOM-drift events with severity and linked video evidence, in one readable feed.
Smart Directory
Versioned document repository that opens CAD and docs in their native tools — never lost, always traceable.
The record layer & integrations
Section titled “The record layer & integrations”Where verified events become one live database you can query any way — plain language, documents, or AI agents over MCP. Deep dive: Core Platform Overview
The Timeline ledger
An append-only, event-sourced log of every approved change — the single source of truth.
Three-pillar data model
Every operation maps to exactly Inventory, Action, or Blueprint — one simple ontology.
Automatic change propagation
One approval updates inventory, the BOM, and assembly instructions everywhere downstream.
Named integrations
Onshape, NetSuite, Jira, and Slack in and out — Matobas joins your stack, it doesn't replace it.